Meta’s Andromeda update rewrote how the ads delivery system selects and ranks ads for auction. It rolled through 2023 and 2024, giving the algorithm significantly more latitude to find conversions outside advertiser-defined audience parameters. Advertisers who tightened audience controls in response saw delivery costs rise and performance drop. Those who loosened audience constraints and invested in creative volume saw performance improve. The accounts that struggled shared one characteristic: they kept running 2021 targeting structures in a 2024 system.
Understanding what Andromeda actually changed is necessary before diagnosing account performance. The previous system rewarded precise audience targeting. A tightly defined Lookalike or layered interest stack told the algorithm exactly who to reach. Andromeda expanded the signals Meta’s AI uses for placement, reducing the informational value of that precision and increasing the informational value of how individual users respond to specific creative at the moment of delivery.
Why narrow audiences hurt performance post-Andromeda
Narrow audiences under 500,000 people restrict the algorithm’s optimization pool rather than improving targeting precision. Meta’s delivery system needs volume to learn which users convert for a specific creative at a specific time. A Lookalike audience of 50,000 people with three interest layers gives the algorithm too small a sample to find meaningful conversion patterns.
The shift is counterintuitive for advertisers trained on the pre-Andromeda system. Tighter audiences used to mean less wasted spend. Post-Andromeda, tighter audiences mean less data for the algorithm, which produces worse optimization, which produces more wasted spend. Broad audiences with high-quality creative started outperforming narrow audiences with mediocre creative in Q4 2023 and the pattern held through 2024.
The practical implication: remove most audience constraints and let the algorithm find converters at scale. Keep only your geographic targeting and any hard exclusions (existing customers, email list). Drop interest layers and most Lookalike constraints. The creative does the targeting now. Users who respond to your creative are your audience, regardless of what interest categories Meta would have assigned them.
Creative as the primary targeting mechanism
Post-Andromeda, creative quality determines which users see your ads and at what cost. Meta’s algorithm scores creative on engagement rate, completion rate (for video), and click-through rate. Creative that earns high engagement gets cheaper delivery. Creative that earns low engagement gets more expensive delivery. The algorithm treats engagement as a signal that the creative is relevant to the users seeing it, and rewards relevance with lower CPMs.
Accounts outperforming benchmarks in 2024 run 20-30 creative variations simultaneously, rotating based on weekly performance data. A creative testing engine, not five static ads refreshed quarterly. The testing infrastructure identifies winning concepts within two to three weeks and scales them before fatigue sets in. The same creative running for four months in a small audience is the most common pattern in underperforming accounts.
The creative formats that perform best vary by category, but the consistent winner for direct response campaigns is video with a specific problem-solution structure in the first three seconds. The first frame determines whether the user stops scrolling. A problem the target buyer recognizes, shown immediately, produces the strongest stop-scroll rates across product categories.
Budget consolidation: the structural change most accounts need
Campaign consolidation is the highest-leverage account change for most advertisers running post-Andromeda. Five campaigns at $100/day each give the algorithm five separate pools of conversion data to optimize. Two campaigns at $250/day each give it two larger pools with faster learning cycles. The math is straightforward: more conversion events per campaign means faster optimization and better audience selection within each campaign. Five campaigns at $100 each, generating 4 conversions per campaign per month, have 4 events each for the algorithm to learn from. Two campaigns at $250 each, generating 10 conversions per campaign per month, learn from 10 events each. The algorithm’s performance is directly proportional to its data volume.
Accounts that consolidate from fragmented structures into two to three campaigns on the same total budget typically see cost per result improve by 15-30% within four to six weeks as the consolidated campaigns accumulate conversion data at higher velocity. The learning phase of the consolidated campaigns is faster, the audience selection improves faster, and the creative testing produces results faster because each campaign sees enough volume to reach statistical significance on creative performance.
What still works, specifically
Advantage+ campaigns give the algorithm maximum latitude for audience, placement, and budget allocation within a single campaign. For e-commerce, Advantage+ Shopping campaigns outperform equivalent manual campaign structures on ROAS in most account categories. For lead generation, Advantage+ campaigns with strong conversion tracking perform well above 50 monthly conversions.
Conversion API implementation alongside pixel tracking closes the attribution gap that iOS 14.5’s App Tracking Transparency created. The pixel alone misses 15-30% of conversion events. CAPI sends conversion signals directly from your server to Meta, bypassing browser-level tracking restrictions. Optimizing without CAPI means training the algorithm on incomplete data, which produces suboptimal audience selection and worse ROAS.
What stopped working
Interest-based audience stacking produces worse results post-Andromeda than broad targeting with the same creative. Manual bidding with strict cost caps underperforms automated bidding in most categories because it restricts the algorithm’s ability to bid dynamically in high-probability conversion auctions. Running many small campaigns at low budgets splits conversion data too thin for any single campaign to optimize.
The campaign consolidation finding is the most actionable change for accounts with fragmented structures. Merging five small campaigns into two larger ones with the same total budget typically produces a 15-30% improvement in cost per result within four to six weeks as the consolidated campaigns accumulate conversion data faster.
The learning phase after consolidation is the period where most advertisers panic and undo the change. When you consolidate five campaigns into two, the new campaigns enter a learning phase where delivery costs temporarily increase and conversion volume temporarily drops. This phase typically lasts 7-14 days. Advertisers who interpret the learning phase as evidence that consolidation is not working revert to the fragmented structure prematurely and never experience the performance improvement the consolidation would have delivered. Set a calendar reminder to evaluate the consolidated campaign structure after 21 days, not after 5. The 5-day read is learning-phase noise. The 21-day read is actual performance. Documenting this expectation in writing before the consolidation goes live, and sharing it with whoever reviews campaign performance, prevents the premature reversion that undermines the majority of consolidation attempts. The written pre-commitment also creates an accountability record: if performance does not improve by day 21, you have clear evidence the consolidation did not work rather than ambiguous data from a half-completed test that was reversed early. Creative strategy becomes the primary competitive lever when audience targeting loosens. How audiences respond to AI-generated content in ads affects Andromeda’s engagement scoring and determines whether AI-assisted creative gets amplified or penalized by the delivery system.